Signals Inbox·July 28, 2026·Frontier AI

Who is OpenAI’s biggest threat?

Google is OpenAI’s biggest overall threat because it can challenge ChatGPT, intercept its most valuable tasks and finance the fight from products people already use. Anthropic is the sharper direct rival, especially in models, coding and paid professional work.

Q1What does “OpenAI’s biggest threat” actually mean?

OpenAI’s biggest threat should be the rival most capable of weakening its whole position, rather than whichever company tops one model ranking this week.

A model can score better than GPT for a few months and leave ChatGPT’s user habit, OpenAI’s company contracts and its access to capital largely untouched. The harder test is whether a rival can take away the tasks that make OpenAI valuable, from everyday questions to coding and company work, then keep doing it for years through better distribution, lower prices or more computing power.

A company can lose several categories and still qualify. Anthropic could hurt OpenAI badly by becoming the default supplier for valuable professional work. Microsoft could gain leverage through the partnership itself. Meta could make everyday AI too cheap to charge for. Yet the biggest overall threat needs several credible ways to win, because OpenAI now operates across consumer subscriptions, enterprise software, developer APIs, coding tools and infrastructure partnerships.

Q2Is OpenAI still leading the AI race today?

OpenAI leads the overall AI market today, although ChatGPT’s scale now carries more of that lead than clear superiority in models, enterprise adoption or infrastructure.

OpenAI says ChatGPT has more than 900 million weekly active users, over 50 million consumer subscribers and more than nine million paying business users. Sensor Tower separately measured ChatGPT at one billion monthly mobile users, making it the fastest application to reach that level. Those definitions differ, but both put ChatGPT well ahead of every standalone AI assistant in established user habit.

The business underneath ChatGPT has also become substantial. OpenAI now generates about $2 billion in monthly revenue, with enterprise products contributing more than 40%. Its APIs process over 15 billion tokens per minute, and Codex has reached roughly three million weekly users. This is a large platform with several revenue engines, not one famous chatbot.

The cleaner lead has gone. Anthropic now edges OpenAI in Ramp’s latest measure of paid business adoption. Claude Opus 5 sits above GPT-5.6 Sol on the current Artificial Analysis index. Google says its model APIs handle more traffic, Gemini has almost as many app users and Google can place AI inside products it already owns. OpenAI remains first overall, but no company now controls every important part of the race.

Where the leading AI companies are strongest today

Q3Has Google already caught ChatGPT with Gemini?

Google has nearly caught ChatGPT in consumer scale, and its wider product reach already gives Gemini more chances to intercept AI tasks than any standalone rival.

Alphabet’s latest results put the Gemini app at 950 million monthly active users, while daily active users have tripled in a year. Sensor Tower measured ChatGPT at one billion monthly mobile users. The methodologies differ, so pretending the gap is exactly 50 million users would be misleading. Gemini has moved from a distant second choice to roughly ChatGPT’s scale.

Google also reaches people outside the Gemini app. AI Mode in Search has passed one billion monthly users, and its query volume has more than doubled every quarter since launch. Ask YouTube recently reached 140 million users on the watch page in a single month. These audiences overlap, so they should not be added together. They still show how often Google can place Gemini-powered answers inside an existing habit.

Google can take valuable work from OpenAI while people continue using ChatGPT. A travel question answered in Search, a document rewritten in Workspace, a video explained on YouTube or a phone task completed through Android never becomes a ChatGPT session. Google can reduce OpenAI’s opportunities one task at a time.

Developers show the same shift. More than nine million developers build with Google’s models each month, and its APIs process around 22 billion tokens per minute, up from 16 billion one quarter earlier. OpenAI has disclosed more than 15 billion. The companies may count usage differently, but Google has reached the same huge scale and is growing faster.

Q4Is Anthropic now beating OpenAI at work?

Anthropic is OpenAI’s strongest direct rival for paid professional work right now, especially where coding, long-running agents and difficult knowledge tasks justify a high bill.

Ramp’s latest index found that 41% of businesses in its dataset paid Anthropic, compared with 39.5% for OpenAI. Ramp tracks spending from more than 70,000 US businesses, with a sample that leans toward American companies using its finance platform. The crossover still deserves attention because Anthropic rose quickly and retained the lead after Ramp updated its methodology.

Anthropic’s own commercial figures point in the same direction. The company says its revenue run rate has crossed $47 billion, up from roughly $9 billion at the end of the previous year. More than 1,000 customers now spend at least $1 million annually, twice as many as less than two months earlier. Company-reported run rates can move sharply, so the exact figure deserves some caution. The direction does not.

Claude Code explains much of the momentum. Anthropic reported more than $2.5 billion in run-rate revenue from the product, with weekly users doubling since the start of the year. Business subscriptions quadrupled, and enterprise customers now generate more than half of Claude Code revenue. Coding gives customers a clear daily test: did the work ship faster, did the agent finish, and did the output survive review?

OpenAI remains larger across work as a whole. It has more than nine million paying business users, enterprise revenue above 40% of total sales and around three million weekly Codex users. Anthropic is the sharper threat because its growth is concentrated in valuable workloads instead of a wide base of paid seats. Claude is already taking business that OpenAI cannot assume will return with the next GPT release.

Q5Does the best AI model decide whether OpenAI wins?

The best AI model attracts users and trials. Its lead can also disappear within weeks.

The latest Artificial Analysis index ranks Claude Opus 5 at 61, ahead of Claude Fable 5 at 60 and GPT-5.6 Sol at 59. The gap between first and third is two points. Different models lead on coding, multimodal work, long tasks, speed or price, which makes a universal winner increasingly hard to defend.

Stanford’s latest AI Index found the same compression over a longer period. The best US and Chinese models have traded places repeatedly, and the gap between them had narrowed to 2.7 percentage points in its latest measurement. The leading closed model was only 3.3 points ahead of the leading open model. Buyers can now find several systems close enough to the best that product design and price often matter more than a small benchmark advantage.

Anthropic shows how a short model lead can become something durable. It turned strong coding performance into Claude Code, with its own revenue and daily workflow. Google does the same through Search, Workspace and Cloud, while OpenAI uses GPT inside ChatGPT, Codex and company agents. A model lead matters most when the company turns it into a habit or contract before the leaderboard moves again.

Q6Are cheap open models crushing OpenAI’s API pricing?

Cheap open and Chinese models are already squeezing OpenAI’s API pricing, while trusted top-tier products continue to command a premium.

The price gap is now large enough to change how companies build their AI systems. OpenAI lists GPT-5.6 Sol at $5 per million input tokens and $30 per million output tokens. DeepSeek’s official pricing puts V4 Pro at $0.435 for uncached input and $0.87 for output. For a workload with equal input and output volume, DeepSeek’s published token price is roughly 27 times lower. The models differ in performance, reliability and support, but that gap gives engineering teams a strong reason to route routine tasks away from the most expensive provider.

Kimi K3 adds another kind of pressure. Its hosted API costs $3 for input and $15 for output, while the model weights can be downloaded and self-hosted. Stanford’s latest report found that the best open model trails the best closed model by only a few percentage points on its aggregate measure. Cost-sensitive companies can now avoid top-tier API prices without accepting obviously weak performance.

OpenAI can still charge more where the full service earns it. Companies pay for reliability, security reviews, tool use, support, stable capacity and the surrounding ChatGPT or Codex workflow. A $30 output rate can be sensible when the model completes a valuable task that cheaper systems fail repeatedly.

The pain lands hardest on generic inference: classification, extraction, summaries, simple support answers and lower-risk coding assistance. Those workloads are becoming easy to route across providers. OpenAI will need to earn more of its margin through products and completed work, not raw access to intelligence.

Published API pricing for selected frontier and challenger models

Q7Is Meta’s free AI a bigger threat to OpenAI than it looks?

Meta is a serious threat to what consumers will pay for basic AI. It poses less danger to OpenAI’s high-value enterprise business than Google or Anthropic.

Meta can distribute AI through WhatsApp, Instagram, Facebook and Messenger, whose combined daily audience recently reached 3.56 billion people. That gives Meta a route into conversations, discovery and content creation without asking users to install a separate assistant. Even modest adoption across those products can make free AI feel like a standard feature rather than a separate subscription.

Meta can afford to be patient. Its latest reported quarter produced $56.3 billion in revenue and $22.9 billion in operating income. The company can fund models and data centers from advertising while using AI to improve engagement and ad performance. OpenAI needs AI products themselves to carry much more of the economics.

The enterprise gap is plain. Meta trails OpenAI, Anthropic and Google in demonstrated business adoption, and it lacks a hyperscale cloud through which companies can buy managed AI services. Open models also spread value across cloud hosts, hardware vendors and independent developers instead of directing all of it back to Meta.

Meta’s immediate effect is cheaper or free everyday AI. Casual questions, image generation and help inside messaging become harder to sell when billions of users receive a capable alternative at no extra cost. Meta becomes a candidate for the biggest overall threat only if it pairs that reach with repeated top-level models and a credible platform for corporate work.

Q8Could Microsoft turn from OpenAI’s partner into its biggest problem?

Microsoft could hurt OpenAI more abruptly than most rivals. Today it remains the partner OpenAI depends on most.

The amended partnership keeps Microsoft as OpenAI’s primary cloud partner. OpenAI products are expected to ship first on Azure when Azure can support them, and Microsoft keeps a non-exclusive license to OpenAI’s models and products through 2032. OpenAI can now serve customers across other clouds, but its revenue-sharing payments to Microsoft continue through 2030 under a capped arrangement.

Microsoft is also making Azure less dependent on any single model company. Its current enterprise pitch tells customers to use the right model for each job, whether that model comes from OpenAI, Anthropic, Microsoft or an open-source developer. That weakens OpenAI’s privileged position inside the cloud that helped establish its enterprise reach.

The tension is straightforward. Microsoft wants Copilot and Azure to own the customer relationship. OpenAI wants ChatGPT and its agent platform to become the interface above office software and cloud infrastructure. Both benefit from the partnership, and both have reasons to stop the other from controlling too much of the stack.

A hostile break would be costly for Microsoft because it owns a major stake, receives revenue share and benefits from early access to OpenAI products. Google gains much more directly when a task moves away from ChatGPT. Microsoft has the most leverage over OpenAI; Google has the clearest incentive to use every advantage against it.

Q9Is xAI close enough to threaten OpenAI now?

xAI has become too well funded and too strong in computing capacity to ignore, but Grok lacks the users and paying business demand needed to rank as OpenAI’s biggest threat today.

Colossus has grown beyond 220,000 Nvidia GPUs, and xAI says the Memphis site is being expanded toward one million. Few companies can build a cluster of that scale so quickly. Grok 4.5 was trained across tens of thousands of newer GB300 GPUs, showing that the cluster now produces regular competitive releases and has moved beyond a one-time demonstration.

Grok 4.5 is also competitive on several engineering benchmarks. In xAI’s published comparison, it came close to GPT-5.5 and Anthropic’s Fable models on Terminal Bench and led some longer software-engineering tests. Provider-run comparisons deserve caution, but Grok has clearly moved past the stage where it could be dismissed as a social-media chatbot.

Paying demand is the missing piece. xAI recently agreed to provide Anthropic with access to Colossus. Renting capacity to a rival makes financial sense, but it also shows where xAI is strongest today: its computing asset is more established than Grok’s customer base. Anthropic brings the enterprise workloads; xAI supplies part of the infrastructure.

xAI could rise quickly if Grok gains a strong developer ecosystem, large business contracts and deeper distribution through X or Starlink. For now, it belongs among the dangerous wildcards with enough resources to catch up, while its pressure on OpenAI remains narrower.

Q10Which OpenAI rival can afford the AI compute race for the longest?

Google has the safest position in the AI compute race because it can finance chips and data centers from existing cash flow while using the same infrastructure across Search, Cloud, YouTube and Gemini.

Alphabet spent $44.9 billion on capital expenditure in its latest quarter, mostly on technical infrastructure. It generated $39.1 billion in operating cash flow during the quarter and ended with $242.5 billion in cash and marketable securities. Google Cloud revenue reached $24.8 billion, up 82%, while operating income rose to $8.8 billion. Its $514 billion cloud backlog gives the company outside customers to help pay for infrastructure that also trains and serves its own models.

OpenAI has assembled extraordinary resources through partners. It raised $122 billion in its latest financing and has secured capacity across Microsoft, Amazon, Oracle, CoreWeave and Google Cloud, alongside Nvidia, AMD, Cerebras and planned custom chips. That diversification reduces the chance that one supplier can stop OpenAI’s growth. It also leaves OpenAI coordinating a large network of companies that own much of the physical stack.

Anthropic follows a similar approach. It has raised $65 billion, uses Amazon Trainium, Google TPUs and Nvidia GPUs, and has added Colossus capacity through xAI. This gives Claude broad access to compute but puts key suppliers on both sides of the commercial relationship. Google competes with Anthropic while selling it infrastructure; Amazon supports Anthropic while building its own AI products.

Google’s advantage is that the same expensive infrastructure can earn money in several places. A TPU can serve Gemini, a Search feature and a paying Cloud customer. A data-center investment can protect advertising revenue while creating cloud sales. OpenAI and Anthropic can keep raising capital while growth remains exceptional. Google can tolerate a longer payoff period and a harsher funding market.

How the leading challengers finance the compute race

Q11Is ChatGPT’s brand strong enough to keep users loyal?

ChatGPT is the strongest AI brand today, but its users are becoming loyal without being exclusive.

Sensor Tower found that ChatGPT became the fastest mobile application to reach one billion monthly active users. OpenAI also reports over 50 million paid consumer subscribers. That gives OpenAI a huge launchpad for new tools: a coding product, shopping feature or agent can reach an existing audience instead of starting from zero.

People are also spreading their AI use across more products. Sensor Tower found that ChatGPT’s share of the audience using leading AI assistants fell below 50%, even as its absolute user count kept rising. People are adding Gemini, Claude, Grok or DeepSeek while keeping ChatGPT.

A person can prefer ChatGPT for general questions, Claude for code and Gemini inside Google products. Switching takes seconds because the same prompt can be pasted elsewhere. A single permanent winner looks less likely than it did two years ago.

ChatGPT’s brand remains a major advantage: it gives OpenAI first consideration, trust from non-specialists and a large pool of paying users. The next challenge is depth. Memory, files, applications, agents and completed transactions need to make leaving inconvenient. Recognition by itself will not stop users from sending their most valuable tasks elsewhere.

Q12Can companies switch away from OpenAI too easily?

Companies can switch models more easily than they can replace a complete AI workflow. OpenAI’s API margins are therefore more exposed than its best-integrated products.

Ramp’s latest data shows how quickly companies can change providers. Anthropic reached 41% paid adoption among businesses in its dataset, while OpenAI stood at 39.5%. Ramp also found that the heaviest AI users tend to pay several vendors instead of committing to one. Companies already buy several models at once, especially when different providers lead on coding, speed, price or data rules.

Cloud platforms encourage the same behavior. Azure, Google Cloud and Amazon let customers compare several model families behind one procurement relationship. Internal routing software can send sensitive work to a private model, coding to Claude, low-cost classification to DeepSeek and general tasks to OpenAI. The employee may never notice which model handled each request.

Full products are harder to replace. A company that has connected ChatGPT Enterprise to internal data, trained staff, approved security controls and built evaluations around its behavior faces real switching costs. Claude Code can develop similar stickiness through repository knowledge and team workflows. The surrounding system creates the attachment.

OpenAI’s safest revenue should come from owning more of the finished workflow. Generic API traffic can move whenever another provider becomes cheaper or slightly better. A well-integrated agent that employees rely on every day is much harder to dislodge.

Q13Which company can attack OpenAI from every direction?

Google is the only rival that can pressure OpenAI across consumer use, enterprise sales, developer APIs, distribution, infrastructure and financing at the same time.

Anthropic is more dangerous in the narrow contest for the best models and high-value professional work. Meta has a stronger route to free consumer distribution. Microsoft has more contractual leverage. xAI can build computing capacity at remarkable speed. Each rival gives OpenAI a real reason to worry. Google is the one that appears in almost every category.

Gemini competes with ChatGPT. AI Mode intercepts questions inside Search. Workspace places Gemini inside paid work. Android, Chrome and YouTube extend the same models into daily habits. Google Cloud sells infrastructure and enterprise agents, while TPUs give Alphabet more control over cost. Search advertising and Cloud profits fund the effort even before Gemini becomes a standalone profit center.

The latest numbers show Google already using those advantages. It reports 950 million Gemini app users, 22 billion API tokens per minute, nearly 90% of the Fortune 100 using Gemini Enterprise and 82% year-over-year Cloud growth. Those figures use different definitions and should not be added together. Separately, they show adoption across several layers of the business, which is much harder for OpenAI to answer than one rival winning one benchmark.

Breadth of the threat posed by OpenAI’s main rivals

Q14Could Anthropic or Microsoft overtake Google as OpenAI’s biggest threat?

Anthropic is more likely than Microsoft to take Google’s place as OpenAI’s biggest threat, but it needs broader distribution and more control over its economics.

Anthropic already has the core of a direct challenge: customers are paying heavily for Claude, its coding product has become a real business and its latest models rank among the best. A lasting lead across several independent enterprise datasets would strengthen the case. The decisive change would be distribution. Claude needs a consumer habit, workplace channel or embedded platform capable of reaching people before they choose ChatGPT or Gemini.

Meta could change the answer by turning free reach into valuable work. A top-tier Meta model, strong corporate platform and credible cloud route would combine billions of users with much better monetization. xAI could rise through a similar sequence if Grok converts its computing base into sustained model leadership and a large developer economy.

Microsoft becomes the most dangerous company if cooperation with OpenAI breaks down and Azure actively shifts customers toward rival models. That would hit OpenAI quickly because Microsoft sits inside its infrastructure, distribution and economics. The partnership’s continuing financial value makes an open rupture less likely than ordinary competitive friction.

Google could still lose its position through poor execution. Its AI products span several names and interfaces, and the company has a history of moving cautiously when new products threaten established revenue. OpenAI can win by shipping a simpler product faster and keeping ChatGPT as the place people deliberately go for difficult work. It has often been better at that job.

Q15Who is OpenAI’s biggest threat?

Google is OpenAI’s biggest threat today. Anthropic is the competitor most likely to beat OpenAI in a direct model or enterprise contest.

OpenAI still leads overall. ChatGPT has more than 900 million weekly users, more than 50 million subscribers and a large business platform. The company is growing quickly enough to fund new products, attract enormous capital and remain the default name associated with generative AI. Google has not displaced that position.

Google has more credible ways to weaken it than any other company. It can catch ChatGPT through Gemini, answer questions before they reach ChatGPT through Search, place AI inside Workspace and Android, compete for developers through its APIs, sell agents through Cloud and lower costs through its own chips. Alphabet can support all of this with advertising cash flow and a profitable cloud business.

Anthropic ranks second and is already more than a challenger. It leads OpenAI in Ramp’s paid business-adoption data, has built a multibillion-dollar coding product and sits at the top of important model rankings. Anthropic is the answer to “Who is OpenAI’s strongest direct rival?” Google wins the broader question because it attacks more of the system around OpenAI.

Meta threatens consumer pricing. Microsoft holds the greatest partnership leverage. Cheap open models threaten API margins. xAI is the fastest-building wildcard. Each leaves a large part of OpenAI’s position untouched.

Google can win while people keep using ChatGPT. More searches, work tasks, developer calls and business agents only need to stay inside Google’s products. That quieter route is already working, which is why Google poses the biggest threat to OpenAI.